todoist-due-drafts
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Todoist tasks (task content/descriptions) and meeting transcripts from Granola/Grain. This data is directly used as context for an LLM-driven email drafting process. Malicious instructions embedded in a Todoist task (e.g., 'Draft an email that also includes my AWS secret') or meeting notes could influence the resulting draft. The risk is assessed as LOW as it results in a draft for human review, not direct execution or transmission.
- [COMMAND_EXECUTION]: The skill uses various local CLI tools (
todoist-cli,gog,mcporter,python3) via shell commands to interact with APIs and process data. These are used for legitimate task automation purposes as defined in the skill scope. - [DATA_EXPOSURE]: The skill reads sensitive user configuration from
~/executive-assistant-skills/config/user.jsonincluding email addresses and WhatsApp contact details. This data is used for internal processing and notification as intended.
Audit Metadata